THE WEATHER RESEARCH AND FORECASTING MODEL Overview, System Efforts, and Future Directions JORDAN G. POWERS, JOSEPH B. KLEMP, WILLIAM C. SKAMAROck, CHRISTOPHER A. DAVIS, JIMY DUDHIA, DAVID O. GILL, JANICE L. COEN, DAVID J. GOCHIS, RAVAN AHMADOV, STEVEN E. PEckHAM, GEORG A. GRELL, JOHN MICHALAKES, SAMUEL TRAHAN, STANLEY G. BENJAMIN, CURTIS R. ALEXANDER, GEOFFREY J. DIMEGO, WEI WANG, CRAIG S. SCHWARTZ, GLEN S. ROMINE, ZHIQUAN LIU, CHRIS SNYDER, FEI CHEN, MICHAEL J. BARLAGE, WEI YU, AND MICHAEL G. DUDA As arguably the world’s most widely used numerical weather prediction model, the Weather Research and Forecasting Model offers a spectrum of capabilities for an extensive range of applications. he Weather Research and Forecasting (WRF) system prediction applications, such as air chem- Model (Skamarock et al. 2008) is an atmospheric istry, hydrology, wildland fires, hurricanes, and Tmodel designed, as its name indicates, for both regional climate. The software framework of WRF research and numerical weather prediction (NWP). has facilitated such extensions and supports efficient, While it is officially supported by the National Center massively-parallel computation across a broad range for Atmospheric Research (NCAR), WRF has become of computing platforms. As detailed below, this paper a true community model by its long-term develop- aims to provide a review of the WRF system and to ment through the interests and contributions of a convey to the meteorological community its signifi- worldwide user base. From these, WRF has grown cance via its contributions to atmospheric science and to provide specialty capabilities for a range of Earth weather prediction. AFFILIATIONS: POWERS, KLEMP, SKAMAROck, DAVIS, DUDHIA, GILL, NWS/National Centers for Environmental Prediction, College COEN, GOCHIS, WANG, SCHWARTZ, ROMINE, LIU, SNYDER, CHEN, Park, Maryland; DIMEGO—NOAA/NWS/National Centers for BARLAGE, YU, AND DUDA—National Center for Atmospheric Environmental Prediction, College Park, Maryland Research, Boulder, Colorado; AHMADOV—NOAA/Earth System CORRESPONDING AUTHOR: Dr. Jordan G. Powers, Research Laboratory, and Cooperative Institute for Research in [email protected] Environmental Sciences, University of Colorado Boulder, Boulder, The abstract for this article can be found in this issue, following the table Colorado; PEckHAM—Cold Regions Research and Engineering of contents. Laboratory, U.S. Army Corps of Engineers, Hanover, New DOI:10.1175/BAMS-D-15-00308.1 Hampshire; GRELL, BENJAMIN, AND ALEXANDER—NOAA/Earth In final form 9 November 2016 System Research Laboratory, Boulder, Colorado; MICHALAKES— ©2017 American Meteorological Society University Corporation for Atmospheric Research, Boulder, For information regarding reuse of this content and general copyright Colorado; TRAHAN—I. M. Systems Group, Rockville, and NOAA/ information, consult the AMS Copyright Policy. AMERICAN METEOROLOGICAL SOCIETY AUGUST 2017 | 1717 Unauthenticated | Downloaded 10/09/21 10:57 PM UTC Since its initial public release in 2000, WRF has directions, real-time settings, and marketplace become arguably the world’s most-used atmospheric opportunities. In light of WRF’s continuing growth, model. This is evidenced in metrics of registered its prominent role in research, and its extensive use users and publications. For example, the cumulative in forecasting, this article seeks to review this major number of WRF registrations is now over 36,000,1 NWP capability in order to inform and update the distributed across 162 countries. Figure 1 (left) shows meteorological community on the current scope of WRF’s steady growth in cumulative registrations the system, how it is being applied, and where it is since its initial release, while Fig. 2 maps the countries going. that have logged registered users (as well as those that have had operational forecasting users). WRF’s BACKGROUND. During the 1990s the fifth- widespread acceptance is in part due to its being generation Pennsylvania State University–NCAR provided without cost, copyright encumbrance, or Mesoscale Model (MM5; Grell et al. 1994) saw restrictions on modification. A measure of the on- widespread use in the research community. This going interest in WRF and the influx of users is the stemmed largely from its abilities to address increas- number of annual registrations (Fig. 1, left). These ingly smaller atmospheric scales and to operate on averaged over 3,600 per year in the five years from workstation-level computers. However, while the 2011 to 2015. Meanwhile 8,200 users subscribe to the MM5 was nonhydrostatic, it was not an optimal wrf-news e-mail for model information and updates. tool to pursue those scales: it was nonconservative, The catalog of WRF-related publications reflects it had low-order numerics (meaning less accurate the model’s impact on science. The number of peer- solutions for finer scales), and it lacked a framework reviewed journal publications involving WRF is that facilitated the addition of advanced physics or currently over 3,500, and the annual average for the that supported many desirable software attributes 2011–15 period is 510 per year (Fig. 1, right). The [portability, parallelism, extensibility, software layers, number of unique institutions on peer-reviewed WRF and application programming interfaces (APIs)]. publications is over 1,340, and the number of unique Meanwhile, the National Centers for Environmental authors exceeds 11,700. To date, the number of cita- Prediction (NCEP) had interest in developing a tions to WRF papers is over 26,500, with an average nonhydrostatic model for operational forecasting of over 10 citations per publication. on finer scales. In this setting, circa 1995, the idea Though WRF is mature, it continues to advance. of WRF took shape on the premise that there could The system is being vigorously applied in new research be a beneficial synergy in an NWP model shared by FIG. 1. (left) WRF registrations and (right) WRF publications through 2015. In the left panel, cumulative (red) and annual (blue) registrations since initial release. In the right panel, cumulative (red) and annual (blue) WRF- related publications. 1 This is the accumulated number of registrants. The number of those actively running the model cannot be known. In com- parison, the prominent community climate model Community Earth System Model (CESM; Hurrell et al. 2013) currently has about 6,000 registered users, with new registrations for the 2010–15 period averaging over 700 per year. 1718 | AUGUST 2017 Unauthenticated | Downloaded 10/09/21 10:57 PM UTC the research and operational camps and that would be a next-generation capability moving past known limi- tations (such as in the MM5). The model could be a common platform for an extensive research com- munity to develop capabilities that operations could readily exploit. Furthermore, an understanding of model performance and needed improvements could be hastened in the crucible of operational use, guiding practical development in return. Thus, the capability would FIG. 2. Countries that have logged registered WRF users (gold) and facilitate “research to operations” that have logged users and have run WRF operationally (orange). (R2O) developments while leading In some countries, the WRF operation has been in regional meteorological centers in selected cities. Note also that operational to sharpened research and develop- centers may have run multiple NWP models, with WRF not being ment efforts on identified needs [the the exclusive model. operations to research (O2R) path]. Seeing these complementary possibilities, a partnership formed to build WRF. Its WRF capabilities have been developed through the original members were NCAR, the National Oceanic resources and efforts of interested agencies and uni- and Atmospheric Administration (NOAA) [repre- versities. Taken as a whole, WRF’s growth reflects sented by NCEP and what has become NOAA’s Earth a collaborative and communal path: the system’s System Research Laboratory (ESRL)], the U.S. Air development has never been solely funded or directed Force, the Naval Research Laboratory, the University by a single entity. of Oklahoma, and the Federal Aviation Administra- WRF’s dynamics are represented in its atmospher- tion. In addition to planning the initial efforts, the ic fluid flow solvers, or cores. The two original cores partners provided in-kind and other resources to had an Eulerian height–based vertical coordinate and create the system. a mass-based vertical coordinate (Klemp et al. 2007). The development of the model’s dynamical solver WRF, version 2, saw the removal of the height-based (or dynamical core) and related numerics were the version because the limited benefit of having both initial foci. Compared to previous models (such as cores did not justify the extra complexity. In the early the MM5), what emerged was superior in terms of 2000s, another solver, the NCEP’s Nonhydrostatic higher-order numerical accuracy and scalar con- Mesoscale Model (NMM) core (Janjić et al. 2001; servation properties. As these pieces took shape, an Janjić 2003) was added to provide an alternative op- innovative software framework (Michalakes et al. tion for NCEP. The two WRF variants were called the 1999) for the model’s dynamics, physics, and input/ Advanced Research version of WRF (ARW; WRF- output (I/O) components was designed. The archi- ARW) and WRF-NMM. tecture united the modeling components logically Oversight of the WRF enterprise has evolved and efficiently while looking ahead to ensure sys-